Distributing Hot-Spot Addressing in Large-Scale Multiprocessors
IEEE Transactions on Computers
High-speed switch scheduling for local-area networks
ACM Transactions on Computer Systems (TOCS)
Highly parallel computing (2nd ed.)
Highly parallel computing (2nd ed.)
Scheduling algorithms for input-queued cell switches
Scheduling algorithms for input-queued cell switches
The Use of Feedback in Multiprocessors and Its Application to Tree Saturation Control
IEEE Transactions on Parallel and Distributed Systems
Deadlock-Free Adaptive Routing in Multicomputer Networks Using Virtual Channels
IEEE Transactions on Parallel and Distributed Systems
Self-Tuned Congestion Control for Multiprocessor Networks
HPCA '01 Proceedings of the 7th International Symposium on High-Performance Computer Architecture
Alleviating Memory Contention in Matrix Computations on Large-Scale Shared-Memory Multiprocessors
Alleviating Memory Contention in Matrix Computations on Large-Scale Shared-Memory Multiprocessors
HPCA '05 Proceedings of the 11th International Symposium on High-Performance Computer Architecture
Efficient Reduction of HOL Blocking in Multistage Networks
IPDPS '05 Proceedings of the 19th IEEE International Parallel and Distributed Processing Symposium (IPDPS'05) - Workshop 9 - Volume 10
Cost / performance trade-offs and fairness evaluation of queue mapping policies
Euro-Par'05 Proceedings of the 11th international Euro-Par conference on Parallel Processing
Scalable alternatives to virtual output queuing
ICC'09 Proceedings of the 2009 IEEE international conference on Communications
An efficient strategy for reducing head-of-line blocking in fat-trees
Euro-Par'10 Proceedings of the 16th international Euro-Par conference on Parallel processing: Part II
Hi-index | 0.00 |
In future interconnection networks, congestion management is likely to become a critical issue owing to increasing power consumption and cost concerns. As congested packets introduce head-of-line (HoL) blocking to the rest of packets, congestion spreads quickly. The best-known solution to HoL blocking, Virtual Output Queues (VOQs), is not scalable at all or too costly when implemented in large networks. In previousworks, we proposed an efficient and costeffective solution, referred to as Destination-Based Buffer Management (DBBM). DBBM groups destinations into different sets, and packets addressed to destinations in the same set are mapped to the same queue. DBBM eliminates most of the HoL blocking (among packets addressed to different sets). It achieves very good results in terms of scalability, throughput, and robustness. However, depending on the distribution of packet destinations, it may introduce an uncertain degree of unfairness among packets mapped on the same queue. In order to overcome this problem we propose the Dynamic DBBM mechanism (DDBBM). DDBBM dynamically eliminates completely the HoL blocking. Performance results show that DDBBM keeps (and in some cases improves) the good results achieved by DBBM in terms of throughput and scalability. Moreover, DDBBM solves the unfairness introduced by DBBM. As an example of applicability, in this paper we show that DDBBM can be applied to InfiniBand with no hardware modification.